Skip to content

Commit

Permalink
added forwards backwards testing
Browse files Browse the repository at this point in the history 
Added forwards backwards testing and now include missingness appropriately

added missingness to diploid LS

added some fixes for flake errors

added missingness to diploid viterbi

changed test_genotype_matching_fb.py

remove stray print

removed caps for bool EQUAL_BOTH_HOM etc

Removed caps for EQUAL_BOTH_HOM etc in Viterbi

removed unused imported function
  • Loading branch information
@astheeggeggs @jeromekelleher
astheeggeggs authored and jeromekelleher committed Jul 5, 2023
1 parent 93cd81f commit 4926a25
Show file tree
Hide file tree
Showing 2 changed files with 133 additions and 69 deletions.
128 changes: 83 additions & 45 deletions python/tests/test_genotype_matching_fb.py
View file
Original file line number Diff line number Diff line change
@@ -1,4 +1,3 @@
# Simulation
import copy
import itertools

Expand All @@ -14,6 +13,8 @@
REF_HOM_OBS_HET = 1
REF_HET_OBS_HOM = 2

MISSING = -1


def mirror_coordinates(ts):
"""
Expand Down Expand Up @@ -411,6 +412,7 @@ def update_probabilities(self, site, genotype_state):
]

query_is_het = genotype_state == 1
query_is_missing = genotype_state == MISSING

for st1 in T:
u1 = st1.tree_node
Expand Down Expand Up @@ -444,6 +446,7 @@ def update_probabilities(self, site, genotype_state):
match,
template_is_het,
query_is_het,
query_is_missing,
)

# This will ensure that allelic_state[:n] is filled
Expand Down Expand Up @@ -561,7 +564,14 @@ def compute_normalisation_factor_dict(self):
raise NotImplementedError()

def compute_next_probability_dict(
self, site_id, p_last, inner_summation, is_match, template_is_het, query_is_het
self,
site_id,
p_last,
inner_summation,
is_match,
template_is_het,
query_is_het,
query_is_missing,
):
raise NotImplementedError()

Expand Down Expand Up @@ -670,41 +680,45 @@ def compute_next_probability_dict(
is_match,
template_is_het,
query_is_het,
query_is_missing,
):
rho = self.rho[site_id]
mu = self.mu[site_id]
n = self.ts.num_samples

template_is_hom = np.logical_not(template_is_het)
query_is_hom = np.logical_not(query_is_het)

EQUAL_BOTH_HOM = np.logical_and(
np.logical_and(is_match, template_is_hom), query_is_hom
)
UNEQUAL_BOTH_HOM = np.logical_and(
np.logical_and(np.logical_not(is_match), template_is_hom), query_is_hom
)
BOTH_HET = np.logical_and(template_is_het, query_is_het)
REF_HOM_OBS_HET = np.logical_and(template_is_hom, query_is_het)
REF_HET_OBS_HOM = np.logical_and(template_is_het, query_is_hom)

p_t = (
(rho / n) ** 2
+ ((1 - rho) * (rho / n)) * inner_normalisation_factor
+ (1 - rho) ** 2 * p_last
)
p_e = (
EQUAL_BOTH_HOM * (1 - mu) ** 2
+ UNEQUAL_BOTH_HOM * (mu**2)
+ REF_HOM_OBS_HET * (2 * mu * (1 - mu))
+ REF_HET_OBS_HOM * (mu * (1 - mu))
+ BOTH_HET * ((1 - mu) ** 2 + mu**2)
)

if query_is_missing:
p_e = 1
else:
query_is_hom = np.logical_not(query_is_het)
template_is_hom = np.logical_not(template_is_het)

equal_both_hom = np.logical_and(
np.logical_and(is_match, template_is_hom), query_is_hom
)
unequal_both_hom = np.logical_and(
np.logical_and(np.logical_not(is_match), template_is_hom), query_is_hom
)
both_het = np.logical_and(template_is_het, query_is_het)
ref_hom_obs_het = np.logical_and(template_is_hom, query_is_het)
ref_het_obs_hom = np.logical_and(template_is_het, query_is_hom)

p_e = (
equal_both_hom * (1 - mu) ** 2
+ unequal_both_hom * (mu**2)
+ ref_hom_obs_het * (2 * mu * (1 - mu))
+ ref_het_obs_hom * (mu * (1 - mu))
+ both_het * ((1 - mu) ** 2 + mu**2)
)

return p_t * p_e


# DEV: Sort this
class BackwardAlgorithm(LsHmmAlgorithm):
"""Runs the Li and Stephens forward algorithm."""

Expand Down Expand Up @@ -737,29 +751,35 @@ def compute_next_probability_dict(
is_match,
template_is_het,
query_is_het,
query_is_missing,
):
mu = self.mu[site_id]

template_is_hom = np.logical_not(template_is_het)
query_is_hom = np.logical_not(query_is_het)

EQUAL_BOTH_HOM = np.logical_and(
np.logical_and(is_match, template_is_hom), query_is_hom
)
UNEQUAL_BOTH_HOM = np.logical_and(
np.logical_and(np.logical_not(is_match), template_is_hom), query_is_hom
)
BOTH_HET = np.logical_and(template_is_het, query_is_het)
REF_HOM_OBS_HET = np.logical_and(template_is_hom, query_is_het)
REF_HET_OBS_HOM = np.logical_and(template_is_het, query_is_hom)

p_e = (
EQUAL_BOTH_HOM * (1 - mu) ** 2
+ UNEQUAL_BOTH_HOM * (mu**2)
+ REF_HOM_OBS_HET * (2 * mu * (1 - mu))
+ REF_HET_OBS_HOM * (mu * (1 - mu))
+ BOTH_HET * ((1 - mu) ** 2 + mu**2)
)
if query_is_missing:
p_e = 1
else:
query_is_hom = np.logical_not(query_is_het)

equal_both_hom = np.logical_and(
np.logical_and(is_match, template_is_hom), query_is_hom
)
unequal_both_hom = np.logical_and(
np.logical_and(np.logical_not(is_match), template_is_hom), query_is_hom
)
both_het = np.logical_and(template_is_het, query_is_het)
ref_hom_obs_het = np.logical_and(template_is_hom, query_is_het)
ref_het_obs_hom = np.logical_and(template_is_het, query_is_hom)

p_e = (
equal_both_hom * (1 - mu) ** 2
+ unequal_both_hom * (mu**2)
+ ref_hom_obs_het * (2 * mu * (1 - mu))
+ ref_het_obs_hom * (mu * (1 - mu))
+ both_het * ((1 - mu) ** 2 + mu**2)
)

return p_next * p_e


Expand Down Expand Up @@ -797,18 +817,33 @@ def example_genotypes(self, ts):
s = H[:, 0].reshape(1, H.shape[0]) + H[:, 1].reshape(1, H.shape[0])
H = H[:, 2:]

genotypes = [
s,
H[:, -1].reshape(1, H.shape[0]) + H[:, -2].reshape(1, H.shape[0]),
]

s_tmp = s.copy()
s_tmp[0, -1] = MISSING
genotypes.append(s_tmp)
s_tmp = s.copy()
s_tmp[0, ts.num_sites // 2] = MISSING
genotypes.append(s_tmp)
s_tmp = s.copy()
s_tmp[0, :] = MISSING
genotypes.append(s_tmp)

m = ts.get_num_sites()
n = H.shape[1]

G = np.zeros((m, n, n))
for i in range(m):
G[i, :, :] = np.add.outer(H[i, :], H[i, :])

return H, G, s
return H, G, genotypes

def example_parameters_genotypes(self, ts, seed=42):
np.random.seed(seed)
H, G, s = self.example_genotypes(ts)
H, G, genotypes = self.example_genotypes(ts)
n = H.shape[1]
m = ts.get_num_sites()

Expand All @@ -819,13 +854,16 @@ def example_parameters_genotypes(self, ts, seed=42):

e = self.genotype_emission(mu, m)

yield n, m, G, s, e, r, mu
for s in genotypes:
yield n, m, G, s, e, r, mu

# Mixture of random and extremes
rs = [np.zeros(m) + 0.999, np.zeros(m) + 1e-6, np.random.rand(m)]
mus = [np.zeros(m) + 0.33, np.zeros(m) + 1e-6, np.random.rand(m) * 0.33]

for r, mu in itertools.product(rs, mus):
e = self.genotype_emission(mu, m)

for s, r, mu in itertools.product(genotypes, rs, mus):
r[0] = 0
e = self.genotype_emission(mu, m)
yield n, m, G, s, e, r, mu
Expand Down
74 changes: 50 additions & 24 deletions python/tests/test_genotype_matching_viterbi.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,8 @@
REF_HOM_OBS_HET = 1
REF_HET_OBS_HOM = 2

MISSING = -1


class ValueTransition:
"""Simple struct holding value transition values."""
Expand Down Expand Up @@ -390,6 +392,7 @@ def update_probabilities(self, site, genotype_state):
]

query_is_het = genotype_state == 1
query_is_missing = genotype_state == MISSING

for st1 in T:
u1 = st1.tree_node
Expand Down Expand Up @@ -423,6 +426,7 @@ def update_probabilities(self, site, genotype_state):
match,
template_is_het,
query_is_het,
query_is_missing,
u1,
u2,
)
Expand Down Expand Up @@ -486,6 +490,7 @@ def compute_next_probability_dict(
is_match,
template_is_het,
query_is_het,
query_is_missing,
node_1,
node_2,
):
Expand Down Expand Up @@ -830,6 +835,7 @@ def compute_next_probability_dict(
is_match,
template_is_het,
query_is_het,
query_is_missing,
node_1,
node_2,
):
Expand All @@ -841,26 +847,28 @@ def compute_next_probability_dict(
double_recombination_required = False
single_recombination_required = False

template_is_hom = np.logical_not(template_is_het)
query_is_hom = np.logical_not(query_is_het)

EQUAL_BOTH_HOM = np.logical_and(
np.logical_and(is_match, template_is_hom), query_is_hom
)
UNEQUAL_BOTH_HOM = np.logical_and(
np.logical_and(np.logical_not(is_match), template_is_hom), query_is_hom
)
BOTH_HET = np.logical_and(template_is_het, query_is_het)
REF_HOM_OBS_HET = np.logical_and(template_is_hom, query_is_het)
REF_HET_OBS_HOM = np.logical_and(template_is_het, query_is_hom)

p_e = (
EQUAL_BOTH_HOM * (1 - mu) ** 2
+ UNEQUAL_BOTH_HOM * (mu**2)
+ REF_HOM_OBS_HET * (2 * mu * (1 - mu))
+ REF_HET_OBS_HOM * (mu * (1 - mu))
+ BOTH_HET * ((1 - mu) ** 2 + mu**2)
)
if query_is_missing:
p_e = 1
else:
template_is_hom = np.logical_not(template_is_het)
query_is_hom = np.logical_not(query_is_het)
equal_both_hom = np.logical_and(
np.logical_and(is_match, template_is_hom), query_is_hom
)
unequal_both_hom = np.logical_and(
np.logical_and(np.logical_not(is_match), template_is_hom), query_is_hom
)
both_het = np.logical_and(template_is_het, query_is_het)
ref_hom_obs_het = np.logical_and(template_is_hom, query_is_het)
ref_het_obs_hom = np.logical_and(template_is_het, query_is_hom)

p_e = (
equal_both_hom * (1 - mu) ** 2
+ unequal_both_hom * (mu**2)
+ ref_hom_obs_het * (2 * mu * (1 - mu))
+ ref_het_obs_hom * (mu * (1 - mu))
+ both_het * ((1 - mu) ** 2 + mu**2)
)

no_switch = (1 - r) ** 2 + 2 * (r_n * (1 - r)) + r_n**2
single_switch = r_n * (1 - r) + r_n**2
Expand Down Expand Up @@ -919,18 +927,33 @@ def example_genotypes(self, ts):
s = H[:, 0].reshape(1, H.shape[0]) + H[:, 1].reshape(1, H.shape[0])
H = H[:, 2:]

genotypes = [
s,
H[:, -1].reshape(1, H.shape[0]) + H[:, -2].reshape(1, H.shape[0]),
]

s_tmp = s.copy()
s_tmp[0, -1] = MISSING
genotypes.append(s_tmp)
s_tmp = s.copy()
s_tmp[0, ts.num_sites // 2] = MISSING
genotypes.append(s_tmp)
s_tmp = s.copy()
s_tmp[0, :] = MISSING
genotypes.append(s_tmp)

m = ts.get_num_sites()
n = H.shape[1]

G = np.zeros((m, n, n))
for i in range(m):
G[i, :, :] = np.add.outer(H[i, :], H[i, :])

return H, G, s
return H, G, genotypes

def example_parameters_genotypes(self, ts, seed=42):
np.random.seed(seed)
H, G, s = self.example_genotypes(ts)
H, G, genotypes = self.example_genotypes(ts)
n = H.shape[1]
m = ts.get_num_sites()

Expand All @@ -941,13 +964,16 @@ def example_parameters_genotypes(self, ts, seed=42):

e = self.genotype_emission(mu, m)

yield n, m, G, s, e, r, mu
for s in genotypes:
yield n, m, G, s, e, r, mu

# Mixture of random and extremes
rs = [np.zeros(m) + 0.999, np.zeros(m) + 1e-6, np.random.rand(m)]
mus = [np.zeros(m) + 0.33, np.zeros(m) + 1e-6, np.random.rand(m) * 0.33]

for r, mu in itertools.product(rs, mus):
e = self.genotype_emission(mu, m)

for s, r, mu in itertools.product(genotypes, rs, mus):
r[0] = 0
e = self.genotype_emission(mu, m)
yield n, m, G, s, e, r, mu
Expand Down

0 comments on commit 4926a25

Please sign in to comment.